Konstantin Georgiev (MSc, BEng (Hons))

Thesis title: Predicting rehabilitation needs and trajectories of older patients

Precision Medicine DTP

Year of study: 2

  • Precision Medicine DTP
  • BHF/University Centre for Cardiovascular Science
  • College of Medicine & Veterinary Medicine

Contact details

PhD supervisors:

Address

Street

Chancellor's Building
49 Little France Crescent
Little France Campus
The Royal Infirmary of Edinburgh

City
Edinburgh
Post code
EH164SB

Availability

  • I'm generally flexible and open to opportunities at the moment, feel free to get in touch by email.

Biography

Background

Ever since my 2nd year as an undergraduate, I have always been interested in applied Data Science and Machine Learning. After completing my BEng in Software Engineering in 2019, I decided to pursue a career in AI, continuing my education at the University of Aberdeen. I managed to earn a place in the MSc DataLab scholarship programme as part of my studies which was incredibly helpful in refining my career path in the field of healthcare. In 2020, I was able to secure an industrial placement with a Glasgow-based healthcare startup (RedStar) as part of my dissertation project. The project was related to Explainable AI methodologies for outcome prediction and analysis of discharge letters for patients in ICUs, which allowed me to take the first step in the healthcare industry. After a successful completion of the project, I decided to continue working at RedStar, gaining domain experience in a variety of different fields involving the optimization of care pathways for diabetes, hip fracture and multimorbidity, before embarking on my PhD. The hands-on experience I managed to achieve in these projects was incredibly valuable and is something I hope to translate in my current research and broaden my understanding.

CV

PDF icon 130349.pdf

Qualifications

MSc graduate in Artificial Intelligence (University of Aberdeen), MSc DataLab alumni

BEng graduate in Software Engineering (Technical University of Varna)

Responsibilities & affiliations

Precision Medicine DTP student

Data Scientist at RedStar AI (https://redstar.ai/)

HighSTEACS, Ageing and Health, AIAI (School of Informatics) research groups

Current research interests

I am currently interested in applied Data Science in the field of geriatric medicine, particularly rehabilitation needs and trajectories of older patients. This includes population analysis on the impact and constraints in treatment of multimorbid patients across different domain areas. By studying these trajectories in greater detail, my aim is to explore how to generate clinically meaningful features derived from timestamped data using process mining techniques (e.g. tracking ward and specialty movements in ICU) and build more efficient predictors for outcomes, such as mortality, readmission and projected recovery. I am also interested in applying explainable AI techniques that can assist in translating these results to healthcare professionals (nurses and physiotherapists) through rich visualizations and treatment recommendations that can minimize healthcare costs and improve quality-of-life of these types of patients.

Past research interests

In general, my research was usually focused on applied Machine Learning techniques for optimizing the decision-making process of end-users, enhancing security and providing interpretability for predictive models. In the field of Computer Vision, I was particularly interested in Visual Object Tracking and Anomaly Detection algorithms for monitoring ongoing conditions within a specific scene. Other than that, I also enjoyed fiddling with algorithms for Generative Modelling, such as GANs, Style Transfer, Image Colorization and Super-resolution for augmenting image and video datasets. In my MSc dissertation project, I also explored Natural Language Processing techniques in healthcare for interpreting free-text notes and explaining suggested clinical outcomes.

Project activity

Abstract:

Frailty and acute disability are growing concerns within healthcare services, many outcomes of which are unpredictable. Healthcare resources that can be allocated to support the rehabilitation of individuals with complex symptoms are finite. Additionally, inactivity during hospitalization often results in increased risk of decline in independence, quality-of-life and can even lead to the acquisition of new disabilities. Untimely intervention and unclear decision boundaries that define the patient condition are typically the cause of this. Current approaches in geriatric medicine do not support personalized care for multimorbid patients. With the recent introduction of Electronic Health Records, there is a major unmet potential, in terms of data-driven approaches for rehabilitation assessment. This data can be utilized to build structured care pathways by identifying key factors that drive improvement or decline in patient health throughout their recovery process and ensure the approach that minimises the patient risk is undertaken. These analyses can then be communicated to professionals using explainable Machine Learning techniques to allow individualised patient profiling and treatment management.

 

The main goal in rehabilitation is to maximise functional recovery after illness. In hospital, this process typically consists of the intervention of multidisciplinary teams of healthcare professionals, such as physiotherapists, nurses, doctors and occupational therapists among others. Thus, the process of communicating expert recommendations for a particular treatment across different specialties can be time-consuming and costly for the recovery of the patient. The acquisition of high volumes of EHR data, particularly timestamped data, brings forward a new opportunity for an automated way of extracting insights on complex interactions between different stages of care. One example is the ability to adapt process mining techniques, commonly used in business modelling, to extract knowledge about hospital episodes. These insights can then be utilised to build multivariate Machine Learning models that will be trained on key factors that truly impact the recovery trajectories on the population-level. The model predictions can be used to track the rehabilitation progress and propose changes in treatment where necessary, according to recovery guidelines. Explainable AI methodologies will also provide support in generating model interpretations and communicating the reasons behind the predicted outcomes (e.g. features that impact the change in recovery rate or the development of a hospital-acquired disability).

 

The main goal of the project is to build tools for structured rehabilitation assessment for patients with frailty. This could consist of:

  • Identification of patterns in treatment history that lead to Hospital-acquired disabilities or decline in condition for particular sets of diseases.
  • Building predictive models that suggest interventions that can minimise the patient route to full recovery.
  • Explainable AI methodologies that can correlate the predicted outcomes to specific interactions during hospital stay.
  • Clear data visualisations presented to both patients and healthcare professionals that increase patient engagement and provide biofeedback of progress to recovery.

 

Current project plan for Year 1:

  • (Sept 2021: Feb 2022) - Training in quantitative and interdisciplinary skills relevant for the rehabilitation project (such as biomedical statistics, population analysis, data and process mining) and completion of compulsory courses.

  • (Feb 2022: May 2022) - Data acquisition, handling and building a representative patient cohort containing a set of patients in rehabilitation wards with markers for frailty.

  • (May 2022: June 2022) - Data analysis of patient state prior to hospitalization (admission patterns, comorbidities, medications, frailty markers), analysis of acute illnesses at point of presentation. Analysis of trends in comorbidities that result in development of specific acute disabilities.

  • (June 2022: July 2022) - Definition of measures of acute hospital frailty and categorisation of frail patient groups; Year 1 review write-up and submission.

  • (July 2022: Sept 2022) - Model development and evaluation for baseline rehabilitation prognosis using the patient cohorts extracted from previous steps. Assessment of clinical outcomes across the population (e.g. trends in mortality, readmission, recovery rate, rate of HAD).

Current project grants

Sir Jules Thorn PhD Scholarship programme

This article was published on